Embedded Files
研究聚焦 / Research Focuses:
研究聚焦 / Research Focuses:
Deep learning-based computer vision & Unregulated pumping well detection in land subsidence areas
Deep learning-based computer vision & Unregulated pumping well detection in land subsidence areas
應用深度學習電腦視覺技術偵測地層下陷區域未納管抽水井
應用深度學習電腦視覺技術偵測地層下陷區域未納管抽水井
Applying Deep Learning Technologies and Street View Images for Recognizing Pumping Wells and Cultivation Types in Land Subsidence (NSTC project MOST 110-2321-B-003-001 - )
Applying Deep Learning Technologies and Street View Images for Recognizing Pumping Wells and Cultivation Types in Land Subsidence (NSTC project MOST 110-2321-B-003-001 - )
Enormous unregulated pumping wells increase the risk of groundwater depletion and environmental disasters, such as land subsidence. However, it remains a great challenge to investigate numerous unregulated wells to manage and model groundwater aquifers. To improve the capacity of groundwater governance, we map and locate private wells using deep learning technologies. We trained and validated convolutional neural networks (CNNs), using Google Street View images. We applied the CNNs to a land subsidence area along the Taiwan high-speed rail, termed the Golden Corridor in Taiwan. The results showed that CNNs can recognize pumping wells with at least 90% accuracy. The testing cases showed their capability to recall all the pumping wells in three road segments along the Golden Corridor. Given the prevalence of unknown private pumping in the Choushui River Fan, our image data-driven computer vision approach not only eases labor-intensive private well investigations but also advances hydrologic understanding for groundwater modeling. We enhance comprehension of unknown sinks and provide their spatial distribution to improve groundwater modeling.
Deep learning neural networks can capture features of pumping wells along streets in subsidence areas.
Ensemble decisions of deep learning neural networks effectively improve accuracy in detecting pumping wells.
Deep learning neural networks provide the spatial distribution of potential pumping of a groundwater system.
AI technology has the potential to improve the capacity building of groundwater governance.
The private pumping wells we detected in a land subsidence area
The private pumping wells we detected in a land subsidence area
The landscape ecological view of vertebrate species richness in urban areas across biogeographic realms.
The landscape ecological view of vertebrate species richness in urban areas across biogeographic realms.
以地景生態學觀點探索不同生物地理分區都市地區脊椎生物多樣性
以地景生態學觀點探索不同生物地理分區都市地區脊椎生物多樣性
The landscape ecological view of vertebrate species richness in urban areas across biogeographic realms.
(NSTC project MOST-107-2410-H-002-245-MY3.)
The landscape ecological view of vertebrate species richness in urban areas across biogeographic realms.
(NSTC project MOST-107-2410-H-002-245-MY3.)
Understanding how the spatial arrangement of remnant green spaces in cities complements biodiversity provides an opportunity for synergy between urban development and biological conservation. However, the geography of urbanization is shifting from Europe and North America to Asia and Africa, and more research is needed for fast-growing regions. To understand how shifting urbanization shapes biodiversity patterns, we analyzed the contribution of landscape factors in explaining vertebrate richness in urban areas across biogeographic realms. We used variation partitioning to quantify and compare the relative importance of landscape factors (composition and configuration) and environmental factors (climate, elevation, and latitude) in explaining vertebrate richness in landscapes with at least a million inhabitants across biogeographic realms. Our results pointed out that in the Indo-Malayan, the Afrotropical, and the Neotropical realm (on average of 16.46%) and China and India (11.88%), the influence of landscape factors on vertebrate richness are significantly higher than that of the Palearctic and Nearctic realms (6.48%). Our findings outline the importance of landscape composition and configuration in shaping biodiversity patterns in regions with fast urban growth during the next two decades, such as Africa, Latin America, and Southeastern Asia. We suggest improving land governance and urban planning to construct eco-friendly landscape structures to mitigate biodiversity loss due to urbanization.
Ecological Modeling & Systematic Conservation Planning.
Ecological Modeling & Systematic Conservation Planning.
生態模擬(物種分布模擬) ＆ 系統性保育規劃
生態模擬(物種分布模擬) ＆ 系統性保育規劃
We developed a spatial optimization-based land-use allocation model, termed the Dynamically Dimensioned Search Landscape Optimization Planning model (DDSLOP), that effectively delivers optimized landscape designs for conserving bird species by enhancing their habitat suitability. We applied the multi-objective DDSLOP to advance landscape design guidelinby taking the trade-off between the suitability indices of species and human habitat into consideration. As such, the study can help identify the potential of synergic use of land resources for urban development and biological conservation.
Page updated
Report abuse